Predicting London Household Energy Use with AI

Working under Prof. Sandeep Shukla, I analysed a dataset of half-hourly energy readings across a representative sample of the Greater London population — 167 million rows of consumption data, household identifiers, and timestamps. The first phase was understanding the data itself: identifying patterns in the time series and handling missing values before any forecasting could begin. From there, I built long-term forecasts by isolating seasonal behaviour in the consumption patterns, then validated model accuracy by interpreting the residual plots — checking that what the model failed to explain was genuinely random noise, not a signal the model had missed.